Every so often, I get sick of basically everything. Walls become suffocating, routine is insufferable, and the city I live in wraps itself against the sky like a cage. So inevitably I duck away and find something to chase (warm faces, the light in autumn, half-formed schemes, etc.), run until I’m dizzy and lost and can’t remember whose couch I’m waking up on or why I crashed there. Weeks later, the sky bruises into swollen dusk, some familiar voice yells for me to come home so I run back into my bed once again, wondering if home is this place more than it is the feeling of staring at an unfamiliar timetable and noticing your heartbeat quicken.

This kinda happened last month so I took a 4 week leave (2 paid, 2 unpaid) from my job to read books, work on open source projects, and couchsurf the East Coast. I spent a lot of rainy days curled up on a friend’s bed in Somerville, MA poking at my laptop, idle afternoons hiding in a corner of the MIT library poking at my laptop, and long electric evenings walking around New York City looking for a place to sit and poke at my laptop. A lot of laptop-poking happened while on “vacation” because I had promised some people that I would give two talks in October, one at SecretCon and one at ToorCon.

Predictably, I put off the ToorCon talk until 2 weeks ago. Also predictably, I started panicking and not sleeping anymore because I said I would show people a new browser fingerprinting technique which did not exist. Somehow, after a lot of head-banging-against-desk, I came up with one that sort-of worked about a week before the ToorCon and actually finished the code right before ToorCon. I named it Sniffly because it sniffs browser history, and also because I was coming down with a cold.

Here’s how Sniffly works:

  1. A user visits the Sniffly page.
  2. Their browser attempts to load images from various HSTS domains over HTTP. These domains were harvested from a scrape of HSTS domains in the Alexa Top 1M. It was really fun to write this scraper; I finally had a chance to use Python’s Twisted!
  3. Sniffly sets a CSP policy that restricts images to HTTP, so image sources are blocked before they are redirected to HTTPS. This is crucial, because If the browser completes a request to the HTTPS site, then it will receive the HSTS pin, and the attack will no longer work when the user visits Sniffly.
  4. When an image gets blocked by CSP, its onerror handler is called. In this case, the onerror handler does some fancy tricks to semi-reliably time how long it took for  the image to be redirected from HTTP to HTTPS. If this time is on the order of a millisecond, it was an HSTS redirect (no network request was made), which means the user has visited the image’s domain before. If it’s on the order of 100 milliseconds, then a network request probably occurred, meaning that the user hasn’t visited the image’s domain.

Here’s a quick demo. It only works in recent Chrome/Firefox versions when HTTPS Everywhere is disabled. The results also turn up a lot of false positives if you are running an adblocker, since ad-blocked domains are indistinguishable from HSTS-blocked domains from a timing perspective. (However, since HTTPS Everywhere domains and ad-blocked domains are mostly the same for every user, they can simply be subtracted out to get more accurate results for users who run these browser extensions.) I didn’t collect analytics on the site, but random testing with several friends showed a ~80% accuracy rate in the demo once browser extensions were accounted for.

For more info, check out the source code, ToorCon slides (pdf), and talk recording. Someone submitted the demo to Hacker News and, to my horror, it was the #1 link for 6+ hours yesterday (!). I feel bewildered that this kind of attention is being granted (again) to random side projects that I do alone in my spare time, but I guess I should take whatever validation I can get right now. It would be sweet if people looked at my work and paid me to hack on interesting stuff for the public so i never had to work a real job again. Maybe someday it’ll happen; until then I’ll prolly hold down a day job and take more fake vacations.

Update: As of March 2016, Sniffly (CVE-2016-1617) has been fixed in the major browsers. Thank you Uncle Google for the bug bounty $$.